Voices

Notes from people who brought a real schema.

We asked alumni to mention a specific module. A few declined to be named. One review is mixed on purpose; the work is not a spa.

Signal Cartography’s funnel honesty week ended our habit of counting a webview load as “started checkout”. Revenue reporting calmed down. The CPO still misses the old hockey stick.

Helen Okonkwo, Head of Product · Manchester

Retention Forensics showed that our “activated” users were just people who dismissed a tooltip. I left slightly embarrassed and with a usable D7 definition.

James · Glasgow

★★★★

Attribution without Theatre did not give us a model we could put on a billboard. It did give us a rule: if a campaign cannot survive a seven-day lag, we do not buy the lag.

Rated on our alumni form · growth pair, fintech

The Session Reconstruction studio assumes you like SQL. I do not. I still finished, because the alternative was another month of “unique users” that were not unique. Would have preferred a slower intro week.

Client in subscription media

Case study · a grocery app that had two activations

Two product colleagues comparing notes at a sunlit table

A United Kingdom grocery app arrived with “first_order” in the product dashboard and “order_paid” in finance. Both were called activation in slides. Pulse Briefing took a morning. Signal Cartography took the following eight weeks.

The repair was unglamorous: a single entry event, a property for fulfilment promise, and a funnel that stopped at paid — not at “tapped the map”. After handover, the weekly pack lost a page. The remaining page could be read aloud without an analyst in the room.

They still argue about substitution rates. That argument is now about operations, which is where it belongs.

Case study · a banking squad after a warehouse move

A meeting room with a whiteboard of journey steps and laptops open

Closed Room, one quarter. The squad had migrated events into a new warehouse and lost session stitching on iOS. Experiment Readouts were being written against a table that double-counted background fetches.

We spent the first month on Session Reconstruction, not on “insights”. The second month, experiment write-ups named the exclusion rules in the first paragraph. A losing onboarding variant was finally killed instead of being “watched”.

Limitation they accepted: we did not restore two years of historic sessions. The Atlas’s capture step has a graveyard. That is cheaper than fiction.